Myth Number One

Protocol Wars have raged in the network design community since the 1970s, or as long as we’ve had packets to switch. In many ways, the current debate over Internet regulation (AKA “net neutrality”) recaps those debates, albeit in a much more foolish way as the loudest voices are the most ignorant. Dana Blankenhorn spills the beans on the article of faith that motivates the regulators:

The idea of absolute top-down control in any area — entertainment, business, technology, politics — is simply non-optimal. Technology has enabled us all to be participants, in any way we wish to be, as well as audience.

That’s heavy, isn’t it? And true! Or not.

“Absolute top-down control” sounds like a really horrible thing, but so does “absolute bottom-up chaos”, because the two most hysterical terms are the first and last. Strip them off and you simply have “top-down”, a pretty respectable concept in many settings. Functional families are managed top-down, and so are businesses, nations, all social groups of animals, etc. So what’s the problem?

Top-down systems are quite often the optimal solution to engineering problems, as they eliminate from the get-go the fundamental problem with edge-managed or bottom-up systems: flapping. If you have four selfish people in a taxi, each of whom wants to get to a different place as quickly as possible, and they can all bark orders at the driver, you may never get anywhere, you’ll simply go around in circles until you run out of gas. All complex systems revolve around compromise, and the top-down control system is in a good position to assess general benefit and ensure that it’s allocated appropriately. It doesn’t scale well, and that’s where modularity and segmentation kick in.

The classic example in networking is the Ethernet. As originally conceived, it was a completely decentralized system, where all decisions about who got to transmit when were made by contention. Each system listened to the wire until it was silent, and then started transmitting. If two systems transmitted at once, they both noticed the collision of their packets and backed-off. After a semi-random delay, they tried again. Eventually everybody got to transmit, but the delays could be substantial. So along comes the Ethernet re-design, where the decision about who gets to transmit is moved up a level, to the switch. Everybody can transmit anytime the switch says they can, and the switch allows several systems to submit packets at the same time, which it sorts out as bandwidth becomes available. Switched Ethernet is centralized, and capable of using 100% of network bandwidth without losing a millisecond to collisions, backoffs, and dead time. It’s the optimal system, and that’s why you use it today.

In this as in many instances, absolute top-control is more optimal than absolute bottom-up chaos. And better than both is moderate top-down management, driven by a policy of the greatest good for the greatest number, or some similar rule.

As an example of engineering fact, it could well be that the best design for a network that permits all people to be performers is for the network’s tubes to be controlled top-down in the interest of fair access. Some at the edges are spammers, virus-makers, and other low-lifes. Can we always count on them to police their own behavior? I don’t think so, and neither do you.

So why do so many grown men still repeat the most idiotic sayings of the scruffy hippies of the 60s?

It’s a great mystery to me.

H/T Doc Searls.

8 thoughts on “Myth Number One”

  1. “Optimal?” Furthest thing from the truth if you want to maximize resources on the medium.

    And yes, from a Nash equilibrium perspective, top down is better than bottom up (but something in between is better than both of them.) That is, a decision by fiat uses resources better than absolute contention for them, and the former is worse than an agreed upon solution.

  2. Mumon,

    Agreements are worthless without a top-down structure the parties to abide by them. Furthermore, agreeing parties usually have little incentive to take into account external effects that they have on non-agreeing parties.

  3. Mumon, is right.

    Network Neutrality is really “net discrimination”; i think that was coined by Doc Searls….

    Nash equalibriums are the furthest thing from telcos/isp/cable MSO’s minds, they are externalizing costs and monetizing bandwidth.

    that’s distinctly different than creating optimal switching topologies and maximizing throughput for end users at a reasonable profit.

    what’s worse is, congress is giving away the store; again the current administration and the “do nothing(for the common good)” congress optimizes profits for business at the expense of the common good.

    No Nash computation is anywhere in sight with BushCo.

  4. The mindreader blognround peers into his crystal ball and sees the wicked profit motive:

    …that’s distinctly different than creating optimal switching topologies and maximizing throughput for end users at a reasonable profit.

    Telcos would die to have one-tenth the profit margin that Google enjoys, dude.

  5. The mindreader blognround peers into his crystal ball and sees the wicked profit motive:

    … After throwing around a bunch of incoherent buzzwords that call into question the posters understanding of the issues technical foundations.


    that’s distinctly different than creating optimal switching topologies and maximizing throughput for end users at a reasonable profit.

    First off, there are only two switching topologies ( defined as Physical and Datalink, Not routing) in wide use today. Stars and Rings. There are meshes as well, but they only work well for wireless… Wired networks are not good candidates for meshing.

    Perhaps you meant technologies? Tony Li (who arguably has done more for the modern Internet than Vint Cerf has.) is working for a wireless startup (tropos I think) , so I think we’re going to see some pretty intense competition from Metro Wireless sooner than expected, something that should help ease peoples fears about “Last Mile Duopolies.”

    Next up, Optimization requires descrimination. All network, and systems architectures have constraints. The first and most important is that the network is reslient to faults and be able to protect itself from harm. This poses some interesting capacity vs. reliability tradeoffs that could be considered “suboptimal” from either a reliability or performance standpoint.

    After that, the requirements get a little muddy, and this is where I think this network theory holy war needs to take a back seat to the realities of modern networking. For example, some things need minimized jitter and delay, not maximized throughput. Should I be required to build two seperate physical networks in order to meet the differing sets of requirements? (An insane proposition by any stretch of the imagination ) Or should I use the same network and have it adapt?

  6. Richard,
    seems we agree on this point.

    is this argument by technical deflection….?
    i’m sorry my generalizing isn’t precise.
    topologies vs technology. me bad.

    my point here isn’t about technical accuracy,
    L2,L3, L4-L7 deep packet processing, load balancing,
    jitter, demod, your right on these points as well as constraints.

    but economically
    who controls the pipe and
    who controls the content?
    what constraints are considered along the way.

    network topologies/technologies, tend to be more efficient than
    pt2pt connections (legacy telco domain). but the idea is to improve
    utilization for the owner of the pipe, content provider, the end user/consumer or all of the above.

    Nash suggests a balance point exists.
    K street isn’t about balance,
    it’s mandating profit, through legistlation.

    that’s what net neutrality bill is about,
    securing the interests of telcos without balancing
    the ‘common interests of americans’ as well as the companies.

    the internet

    my point isn’t what protocol we’re using, whether ARP RARP, with load balance et all et all.. technology isn’t

  7. Actually, we don’t agree. You’re bashing the Telcos for being greedy, while they’re demonstrably less greedy than the pro-neutralist monopolies.

    The point is that you can’t regulate networks by hand-waving, you have to be able to say in precise and unambiguous language what’s good and what’s bad, making sure you don’t cast your net too widely.

  8. is this argument by technical deflection….?i’m sorry my generalizing isn’t precise.topologies vs technology. me bad.

    There is a large body of knowledge and “laws” for the Internet already, and they are primarily technical definitions. Besides RFCs, you have BCPs, (Best Current Practices) which you can use to compare and contrast with the actual operations and configurations of most networks. The problemwith NN legislation, is that it violates these more important operational considerations.

    For example according to the the asinine Markley and Snowe Dorgan, as well as some technically more accurate proposals, BCP38 would not be in compliance. IN otherwords, broadly read these proposals would “break” the Internet.

    technology isn’t the Internet

    Oh thats right, I forgot. It’s magic.

Comments are closed.